Method for an Electronic Controlled Platooning System of Agricultural Vehicles

ICVES 2009 First published in:

Method for an electronic controlled platooning system of agricultural vehicles

X. Zhang, Prof. Dr. M. Geimer, L. Grandl and B. Kammerbauer

automated using electronic control systems [2]. In the past Abstract -This paper presents a method to develop an ten years, many research works have been carried out to electronic controlled platooning system, which will enable an develop an automated agricultural vehicle to replace the autonomous agricultural vehicle to follow a leading tractor labor workforce in the farming operation. In [3] an with a given lateral and longitudinal offset. In our study not only the follow-up motions but also the problems such as automatic steering system was developed to guide a John avoiding obstacle, turning at the end of the field have been Deere 7800 tractor along prescribed straight row courses considered. With the aid of the RTK GPS systems the position with an average error of approximately 2 cm. In [4] a robot of the leading tractor can be obtained with accuracy in the tractor was developed based on RTK-GPS and gyroscope to range of centimeters. The position points of the leading tractor provide navigation information for the path tracking. Such are transmitted by wireless modems to the following vehicle field robot with auto-steering systems are capable of steering continually to provide the target position points for the guidance of the following agricultural vehicle. With the method along target lines automatically, but the application of such of curve fitting a desired path for the following vehicle could be autonomous agricultural vehicles can only be confined to a dynamically created. Based on the target position points and laboratory environment, where obstacles and other safety the path planning, the desired speed and the desired steering related problems could be foreseen. To solve the safety angle of the following tractor are calculated. In order to ensure problems in the real field operations many other high-tech the precise navigation of the driverless following tractor, a sensors have been used to sense the surrounding course tracking controller and a speed controller have to be designed and implemented. In addition to the motion control environment of the farming vehicles. In [5] a machine vision algorithms which could keep the autonomous agricultural based guidance system was demonstrated for an autonomous vehicle following the leading tractor, considerations about agricultural small-grain harvester using a cab-mounted safety and robustness of the whole platooning control system camera. In the recent years laser or laser radar (ladar) have would be issued at the end of this paper. been more and more applied in autonomous vehicles to detect obstacles for the safety reasons. In [6] ladar has been Index Terms - Autonomous agricultural vehicle; Navigation used to navigate a small robot tractor through an orchard and control; Vehicular Safety; Inter-Vehicle-Communication field. However most of the solutions have been successfully realized only in laboratory conditions. Field trials I. INTRODUCTION demonstrated that an automatic guided agricultural vehicle could assist the operator but could not completely replace The demand on increasing agricultural productivity and the operator because of safety considerations. Some the decreasing labor force in the agricultural industry have solutions which have been proved robust in field tests were made the automation of agricultural machinery inevitable. very costly and still a long way from commercialization. On Automation and electrification in the agricultural machinery such a background an electronic controlled platooning have become a trend in the recent years. With the modern system can be regarded as an intermediate step on the road technology such as real-time kinematic (RTK) GPS systems to completely autonomous agricultural vehicles. Because of the accuracy of the positioning can reach 1 to 2 cm per 10 the presence of the operator on one of the agricultural km [1], which makes the precise navigation of agricultural vehicles, the safety problem can be easily resolved without machinery possible. With the introduction of X-BY-Wire consideration of costly sensors and complicated sensor technology in the modern agricultural machinery the fusion algorithm. Thanks to the auto-steering and auto-cruise automatic guidance of land vehicles has become reality and control, which have been implemented as standard technique more and more functions in agricultural vehicles have be in commercial agricultural vehicles, the operator can be relieved from the tedious driving routines and concentrate This work is supported by the Federal Ministry of Food, Agriculture and himself mainly on the supervising of the working processes. Consumer Protection of the German Government. X. Zhang is with the Institute for Vehicle Science and Mobile Machines The primary objective of this paper is to introduce a method at the Karlsruhe Institute of Technology (KIT), Gotthard-Franz-Str. 8, D- to develop an electronic controlled two-tractor platooning 76131 Germany. (phone: 0049-721-6088640; fax: 0049-721-6088609; e- system, which will enable one unmanned tractor to follow up mail: [email protected]). Prof. Dr. M. Geimer is head of the Institute for Vehicle Science and another leading tractor with a given lateral and longitudinal Mobile Machines at KIT. (e-mail: [email protected]). offset. This system can allow one operator to utilize more L. Grandl is with the AGCO Corporation at Marktoberdorf, Germany. than two agricultural machines simultaneously, so that the (e-mail: [email protected]). B. Kammerbauer is with the geo-konzept GmbH at Adelschlag, productivity of the working process will be substantially Germany. (e-mail: [email protected]). improved and the competitiveness of the agriculture

producer will be enhanced. In addition to the follow-up motion which has been demonstrated in many other research works, the obstacle avoidance maneuver and the turn-around of the vehicle at the end of a row will be considered. The safety and robustness of such a platooning system will be issued as well.

II. EXPERIMENTAL EQUIPMENTS AND METHODS Fig. 1 shows one of the experimental agricultural vehicles, which are used to compose the platooning system. The leading vehicle as well as the following vehicle is a 265 kW four-wheel drive Fendt 936 Vario model which is 5.65 m long, 2.75 m wide and 3.37 m high. The equipment used to measure the tractor position of the leading tractor is different from the following tractor. The leading tractor uses a Trimble navigation system, which was mounted by the geo- konzept GmbH. With the AgGPS 252 GPS-receiver attached to the roof of the cab and the 450 radio equipment which receives the real-time kinematic (RTK) signals at 10 Hz data throughput rate, the position accuracy is less then 2.5 cm. Using data from the GPS receiver and internal sensors the position data can be further corrected by the NavController in the cab which can compensate the roll, pitch and yaw movement of the vehicle during measurement. Fig. 2 Hardware architecture of the tractor in the platooning system

In Fig. 3 a method to design a platooning system for two tractors is demonstrated. The leading tractor is driven by an operator and collects its position, heading angle and velocity every 100 ms from satellites. Its position will be corrected by the RTK signal from a reference station and by the navigation controller in the cab. The corrected kinematic information will be read into the AutoBox over a CAN interface. Other status information about the leading tractor Fig. 1 Fendt 936 Vario tractor and its cabin’s inner view will also be collected either periodically or on demand by the AutoBox from the CAN interface. All these information In the following tractor an auto-guide system was already is shaped in desired data set and will be transferred to the installed to measure the position of the vehicle. This system following tractor according to priorities. Once the current is an accessory equipment of the Fendt 936 Vario tractor and position of the leading tractor is obtained, the trajectory of can correct the positioning error caused by the inclination of the leading tractor will be updated using curve fitting. the ground. A gyroscope is also integrated in this auto-guide According to the updated trajectory of the leading tractor, system, so that the positioning can reach the same accuracy the new path segment for the guidance of the following as the Trimble system. Both tractors are equipped with an vehicle can also be obtained every 100 ms. industrial computer, an AutoBox which is the interface between the GPS receiver and the tractor control unit (Fig. lateral offset 2). The industrial computer AutoBox is composed of a Leading tractor PowerPC 750GX processor board running at 1 GHz and several peripheral boards, which can communicate with external equipments over controller area network (CAN) or longitudinal trajectory offset serial interfaces. With the real-time operating system segement of running on the PowerPC the AutoBox performs data leading tractor collection, condition monitoring and control signal computations using software written at KIT.

wireless path segment communication to guide the unmanned vehicle predetermined Following tractor tolerance zone Fig. 3 Schematic diagram of the platooning system for two tractors using GPS and wireless communication.

162

A vital part of an autonomous, unmanned vehicle is Y safety. In such a platooning system, the presence of the operator enhances the safety of the system in unexpected cmaster dangerous situations. To disburden the operator from the X ψ k+1 routine supervising work and assist him by decision making, [x , y ,ψ ] programs doing condition monitoring have been integrated k+1 k+1 k+1 in the software. One of the most important condition cslave ψ ′ monitoring for a platooning system is the distance [x , y ,ψ ] k+1 monitoring. A virtual rectangle safety zone, which always k k k surrounds the unmanned following tractor during its moving, [xk′+1, yk′ +1,ψ k′+1] is conceived to constrain the movement of the tractor and to [x′ , y′ ,ψ ′ ] prevent it from colliding against the leading vehicle. When k k k the following tractor goes beyond the constraints determined + βk [xˆk , yˆk ,ψˆk ] by this safety zone, it will be halted by a real-time program, ∆ψ which will steadily monitor the position of the unmanned + + + vehicle. In order to prevent the unmanned tractor from being [xk , yk ,ψ k ] frequently halted, a predetermined tolerance zone is considered. When the tractor exceeds the inner edge of this tolerance zone, a warning signal will be created to notify the [xk , yk ,ψ k ]:current posture of [xˆk , yˆk ,ψˆ k ]:target point on operator on the leading tractor that the following tractor is the leading tractor the desired path + + + already in the tolerance zone and a suggestion will be given [xk′ , yk′ ,ψ k′ ]:mapping point on [xk , yk ,ψ k ]:current posture of the following tractor according to whether the following tractor is moving too the desired path slow or too fast. For example, if the following tractor has a Fig. 4 Path planning for the following vehicle in the follow-up mode. heavier implement than the leading tractor, it needs more power to overcome the greater tractive resistance, because That means the mapping point locates always in the its implement is deeper in the soil. When the following direction which is perpendicular to the heading of the tractor reaches its power limit, its velocity will be decreased leading tractor. so that it could probably not follow up the leading vehicle. Together with the previous mapping points, the desired path In such a situation the operator on the leading vehicle will cslave for the following vehicle can be updated using curve obtain a suggestion to slow down its velocity, so that the fitting every 100 ms. The instantaneous target point following vehicle can still keep up with the leading one. for the navigation of the following vehicle [ xˆ , yˆ ,zˆ ] will When the following vehicle lags more behind in spite of the k k k be selected on a path segment, which is between its current decreased driving speed of the leading tractor and exceeds position + + + and its desired position on c for the the extern edge of the tolerance zone, both tractors will be [ xk , yk ,zk ] slave halted immediately and a manual inspection must be done. next step [ x′k+1, yk′+1,zk′+1 ]. This path segment between the The width and length the safety zone and the thickness of the current position of the following vehicle and its desired tolerance zone can all be tuned by the operator before the position for the next step can be shaped using a spline system is coupled for platooning. function as:

2 3 y(x) = a0 + a1 ⋅ x + a2 ⋅ x + a3 ⋅ x (2) III. PATH PLANNING

A. Follow-Up With the boundary conditions of gradients as:

The most commonly used motion mode for a platooning y′ (x ) = 1 tan ψ ′ system is the follow-up motion [7]. Fig. 4 shows an example k +1 k +1 k +1 (3) + + how the desired path for the following vehicle is created yk′ (xk ) = /1 tan( ψ k − β k ) from the trajectory of the leading tractor in the follow-up mode. The target point is obtained as the intersection of this spline The solid curve on the left side shows the trajectory of the and the normal of the curve cslave at the point [ xk′ , yk′ ,zk′ ]. leading tractor cmaster . The current position of the leading B. Obstacle Avoidance tractor [x k, y k, ψk] is obtained in equally spaced point in time ( t = 100 ms) from the navigation controller and its The most difficult and perhaps the most crucial issue mapping point on the desired path for the following vehicle related to the safety of an unmanned vehicle is the obstacle avoidance. This problem has been investigated by many [x' k, y' k, ψ'k] will be calculated with a given lateral offset d researchers [8] [9] [10] [11], but few have provided as: absolutely convincing and reliable solutions to prevent the x ' = x + d sin ψ k k k autonomous vehicle from colliding against the obstacle in y ' = y − d cos ψ (1). k k k unforeseeable circumstances. Therefore we have chosen the

ψ k ' = ψ k operator to detect an obstacle which may lay on the desired path of the following vehicle. When the operator perceives a

163

collision risk for the following vehicle, he can guide the left side after the turning. This side change must be taken unmanned vehicle to avoid the obstacle by starting a “Lane- into account before the platooning system starts the follow- Change” maneuver, just with a button press. Fig. 5 shows up motion again. how the following vehicle can take a path with continuous Another turning maneuver is presented in Fig. 6 on the curvature to follow the trajectory of the leading one, when right side, the so-called loop turn [15]. In this turning the operator on the leading vehicle sends a “Lane-Change” maneuver the following tractor remains at the right side of command and return to its previous lane, when it has the leading vehicle, so that the side change of the following avoided the obstacle. In the “Lane-Change” maneuver the tractor doesn’t come into question. normally used third order curve fitting is not sufficient to provide a smooth transition. According to [12], a fifth IV. CONTROLLER DESIGN degree polynomial: A control structure which contains cascade controller with feed forward control is designed to guide the unmanned 2 3 4 5 y(x) = a0 +a1 ⋅ x +a2 ⋅ x +a3 ⋅ x +a4 ⋅ x +a5 ⋅ x (4) tractor along the calculated desired path and to minimize the path error [16]. Fig. 7 demonstrates the structure for the delivers a smooth change between the two transition points speed control which will adjust the velocity of the following (1) and (2). vehicle to keep its distance from the leading tractor constant.

Feed (2) (3) Forward

GF(s)

[xˆk ,yˆk,ψˆk ] vˆk + desired + Position Speed position - controller +- controller (1) (4) ∆vk ∆x ,∆y ,∆ψ [ k k k ] v k Speed sensor fusion [x , y ,ψ ] Fig. 5 Path planning for a “Lane-Change” maneuver to avoid the collision k k k Position sensor fusion

After the operator ensures that the following vehicle has Fig. 7 Structure of the cascade controller with feed-forward control for the already avoided the obstacle, he will send a signal to cancel following vehicle speed. the “Lane-Change”, so that the following vehicle can calculate its path to return to its previous lane with a given In this control structure the current position and speed of lateral offset. the following vehicle are obtained using the sensor fusion algorithm which was programmed in the navigation C. Turn-Around controller. The position and the heading angle of the Fig. 6 shows two most common patterns for the unmanned following vehicle will be compared with the calculated following tractor to turn around [13] [14], when it reaches desired position and the desired heading angle from the path the field end. On the left the example solution for a left hand planning. Using the feed-forward function GF(s) a calculated side turning is presented. bias will be added to the output from the position controller. The sum of both signals will then be compared with the current speed of the following vehicle obtained from sensor fusion. The output of the speed controller is connected to a

(1) (2) CAN interface and can drive the tractor electronic control (3) (4) unit (ECU) to control the mechanical components in the agricultural machine. The structure for the steering angle control is similar to the structure explained above. In this case the position controller will be replaced by a yaw-angle controller, while the speed controller will be replaced by a steering angle controller. Fig. 6 Path planning for the following tractor by turning V. INTER -VEHICLE COMMUNICATION At the point (1) the operator starts the turning of the leading tractor. The following tractor must halt at the A. Hardware mapping point (2) of the point (1) and wait until the operator One of the most important prerequisites for an electronic has finished the turning manually and reached the point (3). controlled platooning system is that the leading and the Then the following vehicle can start its turning following vehicles are connected by a so-called wireless autonomously along the pre-calculated turning path through CAN-bridge, which can collect the data from the controller the mapping points (2) and (4). After the following tractor area network (CAN) bus in one vehicle, transmit it over the has reached the point (4) the whole turning process is air and send the information again to the CAN bus in the completed. A remarkable feature of this turning strategy is other vehicle. Because of the normally large acreage of a that the following tractor changes its relative position to the farm, a wide-coverage mobile communication device with leading vehicle from the right side before the turning to the real-time link ability must be chosen to satisfy the

164

requirements for such an inter-vehicle communication [17]. cruise control system which disburden the operator from the The wireless connection has been successfully established tedious farming routine such as driving and turning, the using two modems working on the 2.4 GHz ISM band and operation safety of the leading vehicle will be enhanced. on the 869 MHz band respectively. The AutoBox, which is Supervision software, which assisted the operator to monitor connected with the wireless modems by serial interfaces the driving and working status of the unmanned vehicle, also (RS-232) and with the vehicle controllers by CAN guaranteed the safety of the following vehicle in normal Interfaces, coordinates the data transmission between CAN circumstances. As a backup of the supervising software the bus and the wireless modems. Table 1 shows the operator can trigger the emergency stopping to halt the specifications of the two wireless modems. Principally we following vehicle immediately in unexpected dangerous need to transmit two types of information between the situations. vehicles: the status information, which must be transmitted A key issue concerning the development of an electronic cyclically, and the command information, which is controlled, safety-related system is to determine the safety transmitted once only when some operation is to be carried integrity level needed for all subsystems. Using the risk out. The 2.4 GHz modem is appropriate for the transmission graph defined in the international standard IEC 61508 [18], of cyclic information because of its larger bandwidth and the severity level of injury and the required performance 100% duty cycle, whereas the 869 MHz modem has a larger levels can be derived when the corresponding subsystem long range but lower throughput (24kpbs × 10%), which is fails. As an example, a risk assessment has been conducted however sufficient for the transmission of acyclic command for the wireless communication used in the platooning information. system (Fig. 8). The break of the wireless communication can cause severe injury (S2) because without the information about the leading vehicle the following one could not be guided correctly and there would be collision danger; the frequency of its exposure to hazard is relative high (F2) because of other interferences in the air; the possibility to avoid the hazard exits by triggering an emergency stopping when the acknowledgement for a successful data

Tab.1 Specification of the communication devices transmission cannot be obtained by the sender in a certain time period. Therefore the risk assessment of the wireless

B. Data Protocol communication will take the red path in the risk graph. The A data protocol, which defines the data type and frame result is the safety integrity level of 2 and a fail-silent format for all the information to be transmitted by the performance is needed for this level. That means the whole wireless modems, has been created to distinguish system must be shut down, when this subsystem fails. Using communication data with different content and different a 1oo2D-architecture [18], the safety integrity level of the priorities. wireless communication can be enhanced to 3. That means the whole system can still work in fail-tolerant mode when a Field Delimiter Frame-ID UTC Longitud Latitude Heading Speed Direction redundant wireless modem is used in this architecture. Bytes 1 1 4 6 6 2 2 2 Data 0xFF 0x02 xxx xxxxxx xxxxxx xx xx xx - Delimiter: Check byte for the start of the frame S1 Frame-ID: Identification for the data frame, 2 stands for the position 1 data from the leading vehicle to the following one P1 UTC: Coordinated Universal Time 1 Longitude: Longitude of the current position of the leading vehicle F1 Latitude: Latitude of the current position of the leading vehicle P2 S2 2 Heading: Angle where the leading vehicle is pointing compared to P1 the true north F2 3 Speed: Velocity of the leading vehicle P2 Direction: Direction in which the leading vehicle are moving S = Severity of injury S1 = negligible S2 = serious Tab. 2 Extraction from the data protocol which defines the type and the F = Frequency or exposure F1 = seldom frame format of the data transmission time to hazard F2 = Frequent to continuous P = Possiblity of avoiding the hazard P1 = Possible

P2 = Scarely possible In Table 2 the position data of the leading vehicle is Fig. 8 Classification of safety-related system in different safety-integrity- defined in a data frame with 32 bytes and with a frame level according to IEC 61508 identifier (frame-ID) of 2. Its frame-ID indicates that this information has a relative higher priority in the whole data According to the different safety-integrity-levels required list. That reflects apparently the fact that the position data is for each subsystem, the signals to be supervised in the very crucial for the safety of the following tractor. Without system are classified in three levels: the warning level with this information the unmanned vehicle could not be guided emergency stopping, the warning level with fault tolerance correctly and there would be collision danger. and the non-warning information level. A user-friendly interface has been developed by the researchers from the VI. SAFETY CONSIDERATIONS AGCO Corporation to show the operator the warning signals when the system goes into fail-silent or fail-tolerant states An emphasis of this work is placed on the safe operation (Fig. 9). of agricultural vehicles. Using the auto-steering and auto-

165

Conference on Precision Agriculture, Minneapolis, MN, June 23-26, 1996, pp.767-778. [4] N. Noguchi, J.F. Reid, Q. Zhang, J.D. Will, K. Ischii, 2001. “Development of robot tractor based on RTK-GPS and gyroscope ” ASAE Paper 01-1195. [5] E.R. Benson, J.F. Reid, Q. Zhang, “Machine vision-based guidance system for agricultural grain harvesters using cut-edge detection ” Biosystems Engineering, vol.86, no.4, pp.389-398, 2003. [6] R. Tsubota, N. Noguchi, A. Mizushima, “ Automatic guidance with a laser scanner for a robot tractor in an orchard ” in Proceedings of the Automation Technology for Off-Road Equipment Conference, Kyoto, Japan, 2004. [7] Z. Zhu, J. Takeda, R. Torisu, J. Chen, Z. Song and E. Mao, “ Control System for Tractor-Platooning ” in Proceedings of the IEEE International Conference on Mechatronics and Automation, August 5-8, 2007, Harbin, China. [8] A. Stentz, C. Dima, C. Wellington, H. Herman, D. Stager, “A System Fig. 9 Screenshots of the User Interface while diagnostic software detects for Semi-Autonomous Tractor Operations ” Autonomous Robots, fault condition. vol.13, no.1, July 2002, pp. 87-104(18) [9] N. Nourani, M. Bosse, J. Roberts, and M. Dunbabin, “ Practical path planning and obstacle avoidance for autonomous mowing ” in VII. CONCLUSIONS Proceedings of the Australasian conference of robotics and In this paper we presented an approach for developing a automation, 2006. [10] V. Subramanian, T.F. Burks and A.A. Arroyo, “ Development of platooning system, which is able to automate an unmanned machine vision and laser radar based autonomous vehicle guidance agricultural vehicle to fulfill some agricultural task, such as systems for citrus grove navigation ” Computers and Electronics in plowing and drilling, cooperatively with another manned Agriculture, vol. 53, 2006, pp. 130-143. leading tractor. Compared with other autonomous [11] V. Subramanian, T.F. Burks, S. Singh, “ Autonomous greenhouse sprayer vehicle using machine vision and ladar for steering control ” agricultural robots which are still far from Applied Engineering in Agriculture, vol. 21, no. 5, pp. 935-943. commercialization, the experimental prototype designed [12] W. Nelson “ Continuous-Curvature Paths for Autonomous Vehicles ” according to our proposal will be able to be converted in a in Proceedings of the IEEE International Conference of Robotics and commercialized product in the near future. A lot of Automation, Scottsdale, AZ, 1989, pp. 1260-1264. [13] N. Noguchi, J. Will, J. Reid, Q. Zhang “ Development of a master- considerations have been dedicated to designing a safe and slave-robot system for farm operations ” Computers and Electronics robust system, so that the developed platooning system can in Agriculture, vol. 44, 2004, pp.1-19. meet the common demands regulated in the safety standards [14] T. Oksanen, A. Visalam, “ Optimal Control of Tractor-Trailer System for electronic controlled vehicles. An interesting and novel in Headlines ” in Proceedings of the Automation Technology for Off- Road Equipment Conference, Kyoto, Japan, 2004. facet of this research is the tolerance zone which constrains [15] D. Bochtis, S. G. Vougioukas, C. Tsatsarelis and Y. Ampatzidis, the movement of the autonomous vehicle. Significant “Optimal Dynamic Motion Sequence Generation for Multiple challenges still lay ahead to determine the dimension of this Harvesters ” in Proceedings of the Automation Technology for Off- tolerance zone and to control the unmanned vehicle Road Equipment Conference, 1-2 September 2006, Bonn, Germany, accurately so that it can always stay in this tolerance zone. pp.33-40. [16] Y. Gao, Q. Zhang “ A Comparison of Three Steering Controllers for Another advantage of our proposal is the supervision of the Off-road Vehicles ” in Proceedings of the Automation Technology for operator as a safety back-up of the system. Preliminary Off-Road Equipment Conference, 1-2 September 2006, Bonn, results from our computer simulation have shown that the Germany, pp.289-301. following vehicle can follow the leading one satisfactorily. [17] N. Murakami, A. Ito, J.D. Will, M. Steffen, K. Inoue, K. Kita, S. Miyaura “Development of a teleoperation system for agricultural vehicles” Computers and Electronics in Agriculture, vol. 63, 2008, path which the leading pp. 81-88. vehicle takes [18] IEC 61508, “Functional safety of electrical/electronic/ programmable electronic safety-related systems”, International Electrotechnical Commission, Switzerland.

desired path path which the following for following vehicle vehicle takes

Fig. 10 Simulation of lane change using the methods described in the path planning.

REFERENCES [1] T. Muhr, P.O. Noack “ Mobile Data Repeaters Enhancing the Availability of RTK Correction Data in the Field ” in Proc. of the ATOE Conference, Sep. 2006, Bonn, Germany, pp.65-69. [2] M. Ehrl, H. Auernhammer “ X-By-Wire via ISOBUS Communication Network” in Proc. of the ATOE Conference, Sep. 2006, Bonn, Germany, pp.227-236. [3] M. O’Connor, T. Bell, G. Elkaim, B.W. Parkinson “ Automatic steering of farm vehicles using GPS ” in Proc. of the 3rd International

166